Please submit both your resume and cover letter.

Why this role exists:

• Drive ambiguous signals to defensible root cause. A KPI regression. A lift number that looks too good. An unexplained cost increase. A model that quietly degraded after an upstream data change. You will trace it across the model, the data pipeline, the auction, and the serving stack — and be right.
• Ship the fix. A new feature, a reformulated objective, a recalibrated model, a changed bidding policy, or a correction upstream in the data. You own it through launch and through the readout afterward.
• Design evaluation that survives scrutiny. Online experiments and offline evaluation in a setting where auctions are non-stationary, treatment and control interfere with each other, and conversions land days late.
• Leave behind methodology, not just results. Tooling and frameworks that let other teams answer the next version of the question without you.
• Communicate to people who will act on it. ML engineers, infrastructure, product, and the account teams sitting across from advertisers.

What we're looking for?

Required

• Ph.D. in computer science, statistics, operations research, economics, or a related quantitative field, plus 2+ years of industry experience — or 5+ years of industry experience solving large-scale ML, optimization, or systems problems without one.
• Strong applied statistics and causal reasoning: experiment design, working with confounded observational data, and the judgment to recognize when a result is not real.
• Fluency in Python and SQL against large datasets (Spark, BigQuery, or equivalent). You can get your own data without waiting on anyone.
• Experience owning a model or algorithm in production — including what happened to it after launch.
• Clear written English. A large share of this role's impact takes the form of a document that changes what a team decides to do.

Strongly Preferred - any of 

• Ads, recommendation, search ranking, marketplaces, or another domain with real-time auctions or bidding.
• Probability calibration, delayed feedback, or selection bias in logged-bandit data.
• CTV, attribution modeling, or incrementality measurement.
• PyTorch or TensorFlow applied to tabular or sequential production data at scale.

Senior vs. Staff

Both are individual-contributor roles. Level is decided at offer, based on demonstrated scope rather than years served.

Senior — in the last 12–18 months you have:

• Taken a project from an ambiguous problem statement to a shipped, measured change in production.
• Made the calls yourself on what to investigate and, just as importantly, what to drop.
• Made at least one teammate measurably better through review, pairing, or mentorship
• Staff — all of the above, plus:
• Led work that spanned multiple teams or systems, where no single team owned the problem.
• Changed what your organization chose to work on, on the strength of analysis you produced.
• Left behind a method, framework, or standard that other people now use by default.
• Represented technical trade-offs directly to non-technical stakeholders, and been trusted to do it.

This role is probably not for you if you are looking for a purely offline research position, or you prefer to hand a model to an engineer and move on to the next paper.